01
Extract aspects
Use supplied aspect terms or the existing spaCy noun-phrase extraction path when the language model is installed.
NLP RESEARCH · MADE INTERACTIVE
Every review has a point of view.
Go beyond a single sentiment score. Explore what people feel about the details that matter — one aspect at a time.
Transparent methods Python ML pipeline Live inference
SENTIMENT, WITH CONTEXT
UNDERSTAND THE WHY, NOT JUST THE SCORE01 — ASPECT INTELLIGENCE
THE WORKSPACE 01 / 03
Put a review under the lens.
Paste real customer feedback. The connected Python service will analyze it using the project’s existing pipeline.
A INPUT
Review input
What did the customer say?
Add comma-separated aspect terms, or leave blank to use spaCy extraction.
Waking the Python service on Render. The first connection can take a little while; this page checks it automatically.
Results are returned by your configured inference service; this interface does not invent predictions.
B AWAITING INPUT
Aspect breakdown
Results from the live inference endpoint
Your insights will appear here.
Run an analysis to see aspect-level polarity, supporting words, and any available trained review-level prediction.
Positive Negative Unknown
INFERENCE STATUS: WAKING SERVICENO FABRICATED METRICS · NO PLACEHOLDER PREDICTIONS
THE PIPELINE 02 / 03
From raw text to useful signals.
One interface. An explicit inference boundary. A Python pipeline that remains the source of truth.
02
Estimate polarity
The current aspect demo uses a transparent proximity lexicon. Its heuristic status is always shown with results.
03
Report honestly
If a trained review-level artifact exists, the API returns its prediction. Otherwise, the UI says it is unavailable.
Research-grade transparency
Read experiment notes Review-level star-rating labels do not validate aspect-level sentiment. This UI keeps those tasks separate and does not display benchmark figures as live predictions.